Running head: FEMALE SEXUALIZATION ON SOCIAL MEDIA 1

Transcription

1 Running head: FEMALE SEXUALIZATION ON SOCIAL MEDIA 1 Income Inequality Not Gender Inequality Positively Covaries With Female Sexualization On Social Media Supplementary Information Appendix Authors: Khandis R. Blake a *, Brock Bastian b, Thomas F. Denson c, Pauline Grosjean a,d, Robert C. Brooks a Affiliations: a Evolution & Ecology Research Centre, and School of Biological, Earth and Environmental Sciences, The University of New South Wales, Sydney NSW 2052, Australia. b School of Psychological Sciences, University of Melbourne, Melbourne VIC 3006, Australia. c School of Psychology, Mathews Building, The University of New South Wales, Sydney NSW 2052, Australia. d School of Economics, The University of New South Wales, Sydney NSW *Correspondence to: Dr Khandis Blake, Evolution & Ecology Research Centre, School of Biological, Earth and Environmental Sciences, The University of New South Wales, Sydney NSW 2052, Australia. k.blake@unsw.edu.au

2 FEMALE SEXUALIZATION ON SOCIAL MEDIA 2 For Supplementary Information, see pp For Statistical Models, see pp

3 FEMALE SEXUALIZATION ON SOCIAL MEDIA 3 Figures Figure S1 represents a non-parametric thin-plate spline generated using the Fields package in R. Thin-plate splines allow the plotting of complex relationships between two independent variables and a dependent variable to be assessed [48]. Unlike the parametric methods we used for hypothesis testing, thin-plate splines do not constrain the kinds of relationship possible. Instead, they apply a smoothing parameter arrived at by a resampling-based generalized crossvalidation (GCV) method, allowing the detection of complex response-surface features if there is good statistical support within the data set for these features. The dependent variable was the count of sexy selfies in each country, log-transformed, and the independent variables were z-score standardized. Once each thin-plate spline was estimated, we plotted the predicted surface as heat maps using the image and contour functions in the Graphics package version of R.

4 Human development FEMALE SEXUALIZATION ON SOCIAL MEDIA 4 Income inequality Figure S1. The relationship between income inequality, human development, and sexy selfies across countries. Note. The heat map is a non-parametric thin-plate spline depicting the prevalence of sexy selfies, where red indicates more selfies. Predictors are z-scores and the dependent variable has been logtransformed to facilitate interpretation.

5 FEMALE SEXUALIZATION ON SOCIAL MEDIA 5 Descriptive Statistics Table S1. Post summary statistics for sexy selfie posts across geographic locations. Location US city US county Nation N (geographies) 5,567 1, N (sexy selfie posts) 10,337 7,680 67,038 Posts per location range posts 0 1,232 posts 1 19,361 posts Posts per location M (SD) 1.86 (15.89) 4.73 (36.69) (1,959.20)

6 Running head: FEMALE SEXUALIZATION ON SOCIAL MEDIA 6 Table S2. Pearson Correlation Table, City Level. Sexyselfies Sexyselfies. restricted uniqueselfies GI.health GI.college GI.reproductive GI.managerial GI.income GI.factor Gini Top5percent 8020ratio EduEmplEarn MedianageF Sexyselfies. restricted unique Selfies GI.health GI.college GI. reproductive GI. managerial GI.income GI.factor Gini Top5percent 8020ratio EduEmplEarn MedianageF Sexratio R.998 **.923 ** * ** **.127 **.128 **.112 ** ** N R.925 ** * ** **.125 **.126 **.110 ** ** N R ** * ** **.153 **.154 **.142 ** ** * N R **.192 ** **.382 **.117 **.057 *.087 ** ** ** * N R **.096 **.121 **.379 ** ** * **.317 ** **.417 ** N R **.071 ** ** ** N R.297 **.615 ** ** ** *.053 * N R.769 ** ** ** ** ** ** N R ** ** ** ** **.226 ** N R.879 **.892 ** ** ** ** N R.681 ** ** ** N R ** ** ** N R.063 * ** N R.158 ** N 1622 Note. ** p <.01. * p <.05. EduEmplEarn = component score of female education, female employment, and female median income.

7 FEMALE SEXUALIZATION ON SOCIAL MEDIA 7 Table S3. Correlation Table, County Level. Sexyselfies Sexyselfies. restricted uniqueselfies GI.health GI.college GI.reproductive GI.managerial GI.income GI.factor Gini Top5percent 8020ratio EduEmplEarn MedianageF Sexyselfies. restricted unique Selfies GI.health GI.college GI. reproductive GI. managerial GI.income GI.factor Gini Top5percent 8020ratio EduEmplEarn MedianageF Sexratio R.998 **.923 ** * ** **.127 **.128 **.112 ** ** N R.925 ** * ** **.125 **.126 **.110 ** ** N R ** * ** **.153 **.154 **.142 ** ** * N R **.192 ** **.382 **.117 **.057 *.087 ** ** ** * N R **.096 **.121 **.379 ** ** * **.317 ** **.417 ** N R **.071 ** ** ** N R.297 **.615 ** ** ** *.053 * N R.769 ** ** ** ** ** ** N R ** ** ** ** **.226 ** N R.879 **.892 ** ** ** ** N R.681 ** ** ** N R ** ** ** N R.063 * ** N R.158 ** N 1622 Note. ** p <.01. * p <.05. EduEmplEarn = component score of female education, female employment, and female median income.

8 FEMALE SEXUALIZATION ON SOCIAL MEDIA 8 Table S4. Correlation Table, National Level. sexyselfies sexyselfies.restricted uniqueselfies sexyselfies instagram Development factor Gender inequality component sexyselfies. restricted unique Selfies sexyselfies instagram Development factor Gender inequality component R **.990 **.826 **.213 * N R **.833 **.212 * N R.991 ** **.226 * N R.833 **.851 ** ** N R.212 *.226 *.246 ** ** ** N R ** ** N Gini Note. ** p <.01. * p <.05.

9 FEMALE SEXUALIZATION ON SOCIAL MEDIA 9 Table S5. Correlation Table, Beauty dataset. Beauty sales GI.health GI.college GI.reproductive GI.income GI.managerial Gini EduEmplEarn medianagef GI.health GI.college GI. reproductive GI.income GI. managerial Gini EduEmplEarn medianagef sexratio R ** ** ** N R **.163 **.106 ** ** N R **.186 **.239 ** **.222 ** ** N R ** ** ** ** N R.359 ** ** ** N R ** ** N R **.078 ** ** N R.324 **.047 * N R.181 ** N 1979 Note. ** p <.01. * p <.05. EduEmplEarn = component score of female education, female employment, and female median income.

10 FEMALE SEXUALIZATION ON SOCIAL MEDIA 10 Table S6. Correlation Table, Clothing dataset. Clothing sales GI.health GI.college GI.reproductive GI.income GI.managerial Gini EduEmplEarn medianagef GI.health GI.college GI. reproductive GI.income GI. managerial Gini EduEmplEarn medianagef sexratio R ** N R **.113 ** * N R **.175 **.325 ** **.271 ** ** N R ** **.082 ** ** ** * N R.394 **.091 **.282 ** * N R.106 **.241 **.071 *.145 ** N R **.055 * ** N R.365 **.080 ** N R.162 ** N 1297 Note. ** p <.01. * p <.05. EduEmplEarn = component score of female education, female employment, and female median income.

11 FEMALE SEXUALIZATION ON SOCIAL MEDIA 11 Table S7. Raw Variable Descriptive Statistics City County Nation Beauty Clothing M SD M SD M SD M SD M SD Gender oppression variables GI (health insurance) 46.26% 7.61% 46.19% 3.44% a a 4.96% 7.07% 44.44% 6.39% GI (college opportunity) 46.87% 4.01% 46.43% 2.38% a a 47.22% 4.19% 52.75% 3.37% GI (reproductive health) a a GI (management occupations) 58.92% 11.40% 61.39% 5.75% a a 59.77% 10.22% 59.88% 10.07% GI (income) 58.23% 7.46% 49.64% 3.00% a a 59.27% 4.93% 58.97% 5.28% Income inequality variables Gini coefficient Top 5% Share :20 ratio Confounders Female education level 91.60% 7.32% 91.36% 4.57% a a 92.41% 5.61% 91.74% 6.29% Female median age a a Female median income $29.3K $9.3K $23.3K $4.6K a a $31.1K $9.1K $30.0K $9.3K Female employment rate 91.9% 4.32% 40.52% 4.82% a a 92.54% 3.41% 92.14% 3.56% Operational sex ratio 52.68% 6.65% 54.03% 4.01% a a 52.38% 6.68% 52.09% 5.45% Urbanization 96.97% 8.64% 58.73% 25.17% a a 97.93% 8.63% 97.78% 9.07% Note. GI = gender inequality. - = data was not available. a = represented using a standardized component score.

12 FEMALE SEXUALIZATION ON SOCIAL MEDIA 12 Table S8. Sexy Selfie Model Fit Statistics Model Nnull Nfinal AICnull AICfinal ΔAIC BICnull BICfinal ΔBIC LLnull LLfinal ΔLL City Model *** City Model *** City Model *** City Model *** County Model *** County Model *** County Model *** County Model *** Nation Model ns Nation Model *** Nation Model *** Nation Model ** Note. ***p <.001, ** p <.01, * p <.05.

13 Running head: FEMALE SEXUALIZATION ON SOCIAL MEDIA 13 Sexy Selfie Method Sexy Selfie Search Terms The sexy and hot words were: sexy, hot, sexydress, hotgirls, sohot, sexygirls, sosexy, hotchicks, sexybabe, sexyladies, sexypic, hotbabe, hotty, sexylegs, sexys, hotness, sexytime, sexybody, hotass, hotwoman, sexyaf, supersexy, hotashell, superhot, sexyashell, hotaf, sexybitch, sexylady, sexyass, sexyasfuck, and hotlegs. The selfie terms were: selfie, selfshot, mirrorselfie, mirrorshot, sexyselfie, and hotselfie. Geolocation Matching For city level analyses, we matched strings in the user s location field to all US cities, villages, boroughs, towns, and census designated places (5,575 city locations) with populations estimated to exceed 5,000 people from the 2015 five year ACS population estimates (the only estimates available when we started analyzing the data) [32]. First, we manually identified locations in the US which shared the same name. We then appended the less populated location with the relevant state or state abbreviation. For example, Arlington Texas (population 291,255) was matched to the term Arlington, whereas Arlington Virginia (population 188,663) was matched to the terms Arlington VA and Arlington Virginia. Where duplicate locations names were in the same state, we appended the least populated location with the location type set by the census (i.e., city, borough, town, or village; otherwise these terms were excluded). For six duplicate location pairs which shared a state and census location type, we retained only the location with the largest population. We also excluded two census designated places named University, Florida due to general ambiguity, leaving n = 5,567 city locations. We then cross-checked each location in this list against the names of (a) all UN recognized countries [49], (b) all cities worldwide with populations exceeding 100K [50], and (c) all counties in the United Kingdom and Canada, again appending cities which shared their name with a county, country, or worldwide city (with a larger population) with their state and state

14 FEMALE SEXUALIZATION ON SOCIAL MEDIA 14 abbreviation. Using macros in Visual Basic, we again match strings in the user s location field with identical strings in the location list described above. All location strings that matched to multiple locations (e.g., the location string NYC Washington DC London matched to three cities) were resolved manually by deletion unless it was clear that only one location was relevant (e.g., I live in Washington DC Go the NYC Nicks! ). The macro matched 10,337 posts to a city, village, borough, census designated place, or town in the USA (range = posts, M = 1.86, SD = 15.89). A random sample of 100 matches showed that 96% of city level locations were correctly matched. We followed this same procedure to match posts to counties in the US. We downloaded a list of all US counties with populations greater than 20,000 people (N = 1,622) [32] and appended duplicate county names with their state name and abbreviation (with the least populated county being appended). We also searched and appended duplicates that matched UK and Canada counties, other countries, or other cities with populations exceeding 100K (as in city level matching). We then aggregated all city posts into their respective counties [51], discarding 2,610 posts from 111 cities which were under the jurisdiction of multiple counties. This procedure matched 7,643 posts from city locations into their respective counties. We then re-searched user location field strings for county matches, geolocating an additional 37 posts. The number of posts across the 1,622 counties ranged from 0 1,232 posts (M = 4.74, SD = 36.69). To match posts to nations worldwide, we utilized a list of all countries recognized by the UN (N = 193 countries) in English, French, Spanish, Italian, and German languages [52]. Added to this list were the names of all cities worldwide with populations exceeding 100K [50] (N = 3,712 cities) and all counties, provinces, states, territories, regions, or districts of the 10 countries with the greatest percentage of Twitter users [53]. We again resolved duplicate names by appending the city with the smallest population with the country name then the country ISO abbreviation. We excluded worldwide locations with less than four characters due to the

15 FEMALE SEXUALIZATION ON SOCIAL MEDIA 15 likelihood of inaccurate matches (N = 29 cities). We then matched user location field strings to identical strings in this list, aggregating matches to the nation level. This procedure matched 68,496 posts to 182 countries worldwide. A random sample of 100 matches showed that 99% of user locations were correctly identified. Although we matched location fields to country names in non-english languages, we tracked sexy selfie search terms in English only, meaning that countries which did not post frequently in English were under-represented. To account for this bias, we measured the frequency of English language posts on Twitter in a 30 second window every 30 minutes for 24- hours (N = 44,492 tweets). We determined the national location of these posts following the same procedure outlined above, then aggregated matches to determine the frequency of English tweets for each nation. Seventy of the 193 UN-recognized countries had zero English posts and were excluded from further analyses; we also excluded 10 countries for whom a Gini coefficient could not be computed due to missing data, leaving n = 113 countries (posts ranged from 1 19,361 posts, M = , SD = ). Thirty-eight percent (38.1%) of these countries were classified by the UN as Very High Human Development Countries; 30.1% were High Human Development Countries; 20.4% were Medium Human Development Countries, and 11.5% were Low Human Development Countries. Sexy Selfie Measures Except where stated otherwise, city and county level measures were collected from the 2016 US Census Bureau five year estimates from the American Community Survey (ACS) [32]. Nation level income inequality was collected from the UN [26] or The World Bank [54] (with preference given to the source which yielded the most recent estimate). Where income inequality estimates from the UN and The World Bank were unavailable, we consulted the Central Intelligence Agency World Factbook [55]. The Gini coefficient was our measure of income inequality. At the city and county level, the measure was computed based on households. At the

16 FEMALE SEXUALIZATION ON SOCIAL MEDIA 16 nation level, coefficients were based on households or individuals (data from the UN is based on both) or households only (data from the CIA and Worldbank). We operationalized gender inequity in US city and US counties as follows. For the reproductive health dimension, we measured the percent of people without health insurance who were women and the adolescent fertility rate (births per year old women). For the empowerment dimension, we measured the percent of people who had achieved some college education who were men and the percent of people in management occupations who were men. The percent of women holding parliamentary seats was not applicable for city and county level analyses and we chose to measure education at the college level rather than secondary level to avoid ceiling effects. For the labor dimension, we measured the percent of the combined male and female median income attributed to men. For all inequality measures, higher scores reflected more gender inequality. The education variable reflected the percentage of women who had achieved more than a year 10 education and the age variable was the median female age. The income variable measured the median earnings for the female civilian population aged 16 years and over. The employment variable was the percent of women aged 16 years and over who were employed. We entered all variables except age into a principal components analysis for city, county, beauty salon, and clothing store data. The KMO statistics and Bartlett s test of sphericity indicated that the data were adequate for component analysis in all cases (KMOs >.60; χ2(3) > , ps <.001), extracting one component which accounted for >60% of the variance. We calculated the operational sex ratio by determining the percent of unmarried year old people who were male. Urbanization reflected the percentage of the local area that was urban (the most recent estimates were from the 2000 US Census Bureau ACS). Human development & English-language posting frequency component analysis The 113 nations used in our analyses varied widely on their level of human development. Because less developed nations have poorer access to telecommunication technologies and access

17 FEMALE SEXUALIZATION ON SOCIAL MEDIA 17 to the Internet is associated with gendered preferences [56], we controlled for differences in human development between countries. We measured gross domestic product per capita, median age, life expectancy, urbanization, population, access to the internet, and the Human Development Index (HDI) at the nation level. The HDI is a composite index measuring average achievement in three basic dimensions of human development a long and healthy life, knowledge, and a decent standard of living [26]. We entered these variables and the percent of English-language Twitter posts into a principal components analysis with varimax rotation. The KMO statistic and Bartlett s test of sphericity indicated that the data were adequate for component analysis (KMO =.86; χ2(28) = , p <.001) and the analysis extracted two components which accounted for 75.88% of the variance. Component 1 (59.43%) comprised the HDI, gross domestic product per capita, median age, life expectancy, urbanization, and internet accessibility and thus appeared to represent general human development. Component 2 (15.00%) comprised population and percent of English posts and thus appeared to represent social media posting volume. Variable loadings ranged from on the respective components and from -.14 to.17 on the alternate component. We saved component scores using the regression method, controlled for Component 1 scores in all nation analyses, and offset all nation models by Component 2 scores to account for social media volume effects on sexy selfie frequency. Cross-national gender inequality component analysis We measured cross-national gender inequality via the physical insecurity, inequality in family law/practice, and government framework for gender inequality subscales from the WomanStats Database [33]. The physical security scale examines laws and practices amongst nations regarding domestic violence, rape and sexual assault, marital or family rape, military and war-related rape, and honor killings/femicide. The scale is ordinal and coding is explained in Table S9. The inequity in family law/practice scale captures how inequitably family law is

18 FEMALE SEXUALIZATION ON SOCIAL MEDIA 18 conceptualized according to gender. It encompasses laws and practices regarding marital rape, age at first marriage, whether women must consent to marry, presence of polygyny, abortion law, attitudes and laws regarding gendered aspects of divorce, and women s property rights. The scale is ordinal and coding is explained in Table S10. The Government Framework for Gender Equality Scale utilizes three subscales that identify a general framework for feminist action: (1) the legal declaration of gender equality, (2) the existence of a gender equality action plan, and (3) a commitment to the international gender equality framework of the Convention on the Elimination of all Forms of Discrimination Against Women. These three subscales together produce a scale ranging from 0 7, capture legislative, practical, and international determinants of feminist government policies. The scale legend is in Table S11. We entered the three scales into a principal components analysis with varimax rotation. The KMO statistic and Bartlett s test of sphericity indicated that the data were adequate for component analysis (KMO =.61; Bartlett s χ2(3) = 95.33, p <.001) and the analysis extracted one component which accounted for 67.68% of the variance. Variable loadings ranged from and we saved the standardized component score using the regression method. This component was strongly correlated with the Gender Inequality Index (GII) from the UN, r(107) =.78, p <.001, which we did not include due to high collinearity with the human development component, r(107) = -.86, p <.001, VIF = Nevertheless, our results replicate using the UN GII.

19 FEMALE SEXUALIZATION ON SOCIAL MEDIA 19 Table S9. Physical Security of Women Scale Scale Scale legend 0 There are laws against domestic violence, rape, and marital rape; these laws are enforced; there are no taboos or norms against reporting these crimes, which are rare. There are no honor killings or femicides. a 1 There are laws against domestic violence, rape, and marital rape; these laws are generally enforced; there are taboos or norms against reporting these crimes (or ignorance that these are reportable crimes), which crimes are not common. Honor killings and femicides do not occur. 2 There are laws against domestic violence, rape, and marital rape; these laws are sporadically enforced; there are taboos or norms against reporting these crimes (or ignorance that these are reportable crimes), which are common. Honor killings and/or femicides are quite rare, occurring only in small pockets of the population, and are condemned by society. 3 There are laws against domestic violence, rape, but not necessarily marital rape; these laws are rarely enforced; there are taboos or norms against reporting these crimes (or ignorance that these are reportable crimes), which affect a majority of women. Honor killings and/or femicides may occur among certain segments of society but are not generally accepted within the society. 4 There are no or weak laws b against domestic violence, rape, and marital rape, and these laws are not generally enforced. Honor killings and/or femicides may occur and are either ignored or generally accepted. Note. a Femicide refers to the targeting killing of women (practices that sanction murder where the overwhelming proportion of victims are female; e.g., witchcraft killings). b An example of a weak law is the need for four male witnesses to prove rape occurred. Table S10. Inequity in Family Law/Practice between Men and Women Scale Scale Scale legend 0 Legal age of marriage is at least 18, and most (80%+) marry over that age. Marriages younger than 16 are virtually unheard of. Polygyny is illegal and extremely rare. Women are free to choose their spouse. Women know their rights to consent and divorce and are free to exercise those rights without fear of reprisal. Marital rape is illegal and actively prosecuted. Women and men have equal rights to divorce. Woman can inherit property upon the death of a parent or upon divorce. Abortion is safe and legal and not imposed by the state on women (i.e. forced abortions are not an issue). 1 Legal age of marriage is 16 or higher and most (80%+) marry over age 16. Polygyny is illegal and uncommon. Women are free to choose their spouse. Women know their rights to consent and divorce and are free to exercise those rights without fear of reprisal. Marital rape is illegal. Women and men have equal rights to divorce. Woman can inherit property, but laws tend to favor men in property rights, including asset division after divorce. Abortion is legal (although may not be available on demand for the asking). 2 Legal age of marriage is 16 or higher, but girls marrying younger are common (up to 25%). There is often an age difference between the legal age of marriage for men and women, such that girls are allowed to marry at younger ages than males.

20 FEMALE SEXUALIZATION ON SOCIAL MEDIA 20 Polygyny is legal but unusual (<5% of women). Girls may not have full rights to choose their spouse. Women may or may not know their rights to consent and divorce. Marital rape may be illegal, but is not prosecuted and practice often allows it. Generally speaking, the grounds for divorce for men and women are the same, although there may be exceptions (i.e., exempting infidelity on the part of the male, or infertility on the part of the female). Divorce laws systematically favor men, and women do not have equal rights in child custody matters. Abortions may be restricted, but there are many reasons for permission to be given, including financial reasons. 3 Legal age of marriage is 15 or lower, but girls marrying younger are common (between 25-50%). Age discrepancies in the average age of men and women getting married is often greater than 7 years or more, with women often averaging less than 15 years old at time of marriage. Polygyny is legal and not uncommon (>5% but less than 25% of women). Girls often cannot chose their spouse. Although obstacles exist that force women to meet a higher standard of justification than men, women can seek divorce but are generally unaware of that right. Women in certain areas of in certain ethnic or religious groups may either be unaware of their rights to consent in marriage and to divorce, or may fear reprisals if they exercise those rights; such rights may be very limited. Marital rape is not acknowledged in law. Divorce laws systematically favor men, and women do not have equal rights in child custody matters, or in inheritance law. Abortions are severely restricted to cases where the life of the mother is at risk, possibly also rape and incest. 4 Legal age of marriage does not exist or allows girls younger than 12 to marry. Girls commonly (more than 25%) marry around the age of 12 or even before puberty. Women are rarely asked for consent before marriage, and women are often forced to marry much older men in this way. Polygyny is legal and common (>25%). Women must overcome tremendous legal obstacles to sue for divorce, while men can seek divorce for many reasons. Women may be unaware of their right to give consent in marriage or to divorce their husbands, may not legally possess such rights, or may feel that the exercise of those rights would bring dire physical or social consequences. Women are not awarded custody or inheritance. Marital rape is not illegal. Abortions are illegal (you may also take cases where states impose abortions on women, i.e., forced/coerced abortions). Table S11. Government Framework for Gender Equality Scale Scale Scale legend 0 1 Strong policies across all three dimensions (law, action plans, CEDAW) 2 3 Strong policies exist on most, but not all, dimensions 4 5 Gender equality policies may exist, but are inadequate on more than one dimension 6 7 No or very weak policies on gender equality across all three dimensions

21 FEMALE SEXUALIZATION ON SOCIAL MEDIA 21 Gender Matching Procedure Though the majority of users had female names, we did not exclude users with male names (or users where gender was indeterminable). To approximate gender for all posts, we manually checked 1,500 randomly selected posts and manually identified user gender and whether they posted photos mainly of themselves or of others. This check identified that 62% of users were women and 71% of users shared photos mainly of themselves. We then used the R package GenderizeR [57] to approximate user gender for each post in the dataset. GenderizeR uses a dataset of 216,286 distinct names across 79 countries and 89 languages to predict gender from first names extracted from text corpuses. We controlled for the accuracy of prediction by setting the counts of first names in the genderize.io names database to 100 and by setting the probability level to >0.60. This process ensured that only frequently cited names with a greater probability of being one gender or another were included. To determine whether posts were selfies or posts of the other gender, we analyzed the frequency of gender-related keywords in the post text (to determine the gender of the subject photographed in the post). We operationalized references to women via the inclusion of the words "girl", "lady", "woman", "chick", "ladies", "babe", "female", "women", "babe", "she", "wife", "mother", " mom", "milf", and " her". The equivalent male words were "male", " boy", " man", "men", gay", "he ", "husband", "father", " dad", "guy", and "his". We split all posts by user gender to determine people posting selfies or people posting photos of the other gender. We found that when men were posting sexy selfies, 87% of them were posting (or re-posting) selfies of women and not of themselves (reposts were more common, comprising 54% of men s posts). By contrast, 90% of women were posting selfies of women (presumably themselves) and not of men. In total, just over three quarters of all posts entailed women posting genuine selfies, and men (and very occasionally, women) reposting them.

22 FEMALE SEXUALIZATION ON SOCIAL MEDIA 22 Model Appendix Table of Contents SUMMARY OF DATA ANALYSIS STRATEGY MODELS REPORTED IN MANUSCRIPT City Models City Model City Model City Model City Model County Models County Model County Model County Model County Model Nation Models Nation Model Nation Model Nation Model Nation Model Beauty Models Beauty Model Beauty Model Beauty Model Beauty Model Clothing Models Clothing Model Clothing Model Clothing Model Clothing Model ROBUSTNESS CHECKS Models with Alternative Measures of Gender Inequality City Models with Single GI Predictors City Model 3 with GI.health City Model 4 with GI.health City Model 3 with GI.reproductive City Model 4 with GI.reproductive City Model 3 with GI.college... 53

23 FEMALE SEXUALIZATION ON SOCIAL MEDIA 23 City Model 4 with GI.college City Model 3 with GI.managerial City Model 4 with GI.managerial City Model 3 with GI.income City Model 4 with GI.income County Models with Single GI Predictors County Model 3 with GI.health County Model 4 with GI.health County Model 3 with GI.reproductive County Model 4 with GI.reproductive County Model 3 with GI.college County Model 4 with GI.college County Model 3 with GI.managerial County Model 4 with GI.managerial County Model 3 with GI.income County Model 4 with GI.income Beauty Models with Single GI Predictors Beauty Model 3 with GI.health Beauty Model 4 with GI.health Beauty Model 3 with GI.reproductive Beauty Model 4 with GI.reproductive Beauty Model 3 with GI.college Beauty Model 4 with GI.college Beauty Model 3 with GI.managerial Beauty Model 4 with GI.managerial Beauty Model 3 with GI.income Beauty Model 4 with GI.income Clothing Models with Single GI Predictors Clothing Model 3 with GI.health Clothing Model 4 with GI.health Clothing Model 3 with GI.reproductive Clothing Model 4 with GI.reproductive Clothing Model 3 with GI.college Clothing Model 4 with GI.college Clothing Model 3 with GI.managerial Clothing Model 4 with GI.managerial Clothing Model 3 with GI.income... 87

24 FEMALE SEXUALIZATION ON SOCIAL MEDIA 24 Clothing Model 4 with GI.income Models with GI Factor City Model 3 with GI Factor City Model 4 with GI Factor County Model 3 with GI Factor County Model 4 with GI Factor Beauty Model 3 with GI Factor Beauty Model 4 with GI Factor Clothing Model 3 with GI Factor Clothing Model 4 with GI Factor Models with Alternate Measures of income Inequality Top 5 Percent Models City Model 2 Top 5 Percent City Model 3 Top 5 Percent City Model 4 Top 5 Percent County Model 2 Top 5 Percent County Model 3 Top 5 Percent County Model 4 Top 5 Percent :20 Ratio Models City Model 2 80:20 Ratio City Model 3 80:20 Ratio City Model 4 80:20 Ratio County Model 2 80:20 Ratio County Model 3 80:20 Ratio County Model 4 80:20 Ratio Models with Additional Pornography Exclusions Models Excluding Pornography Keywords City Model 1 with Additional Pornography Exclusions City Model 2 with Additional Pornography Exclusions City Model 3 with Additional Pornography Exclusions City Model 4 with Additional Pornography Exclusions County Model 1 with Additional Pornography Exclusions County Model 2 with Additional Pornography Exclusions County Model 3 with Additional Pornography Exclusions County Model 4 with Additional Pornography Exclusions Nation Model 2 with Additional Pornography Exclusions Nation Model 3 with Additional Pornography Exclusions

25 FEMALE SEXUALIZATION ON SOCIAL MEDIA 25 Nation Model 4 with Additional Pornography Exclusions Instagram-Only Models Nation Model 1 with Instagram-Only Nation Model 2 with Instagram-Only Nation Model 3 with Instagram-Only Nation Model 4 with Instagram-Only Models with Unique Users Only City Model 1 with Unique Users Only City Model 2 with Unique Users Only City Model 3 with Unique Users Only City Model 4 with Unique Users Only County Model 1 with Unique Users Only County Model 2 with Unique Users Only County Model 3 with Unique Users Only County Model 4 with Unique Users Only Nation Model 1 with Unique Users Only Nation Model 2 with Unique Users Only Nation Model 3 with Unique Users Only Nation Model 4 with Unique Users Only Models with Posts from Women of Women City Model 1 from Women of Women City Model 2 from Women of Women City Model 3 from Women of Women City Model 4 from Women of Women County Model 1 from Women of Women County Model 2 from Women of Women County Model 3 from Women of Women County Model 4 from Women of Women Nation Model 1 from Women of Women Nation Model 2 from Women of Women Nation Model 3 from Women of Women Nation Model 4 from Women of Women Models with Restrictions on City Size City Model 1 with Cities>20K City Model 2 with Cities>20K City Model 3 with Cities>20K City Model 4 with Cities>20K

26 FEMALE SEXUALIZATION ON SOCIAL MEDIA 26 Models with Urbanization then Income Inequality City Model with Urbanization then Income Inequality County Model with Urbanization then Income Inequality Models with only non-weird countries Model 1 with only non-weird countries Model 2 with only non-weird countries Model 3 with only non-weird countries Model 4 with only non-weird countries

27 FEMALE SEXUALIZATION ON SOCIAL MEDIA 27 SUMMARY OF DATA ANALYSIS STRATEGY Data analyses were conducted in R using both Windows and Linux operating systems and the packages glmmadmb, DescTools, AER, and lme4. We evaluated each model iteratively as soon as it is modified, to quantify and qualify the fit improvement [58]. Below is our approach for every count model. We followed the same approach for linear models (but with a gaussian distribution), excluding Steps 1 and We tested the suitability of poisson versus negative binomial distributions, retaining the model with the smallest AIC fit. All poisson models were significantly over-dispersed and none were used. 2. We then compared negative binomial models against zero-inflated negative binomial models using Vuong s test and a comparison of AIC values, to see if excess zeros warranted a dual-approach analytic strategy. All negative binomial models provided superior fits. 3. We added a random intercept and retained it when its inclusion warranted a small decrease in AIC units (approximately Δ2-AIC units). We then added random slopes for all fixed effects, again retaining them when their inclusion warranted a small decrease in AIC units. 4. We excluded standardized Pearson Residuals > ± 2.96, to account for model outliers. For the vast majority of models, no more than 2% of cases were excluded as outliers. Where more than 2% outliers were excluded, we note this below the model output. 5. We compared this final model against the model with no predictors (the null model) using a formalized likelihood ratio test. 6. We evaluated collinearity using Variance Inflation Factors. No predictor yielded VIFs greater than 2.0, and most were below 1.50.

28 FEMALE SEXUALIZATION ON SOCIAL MEDIA 28 City Models City Model 1 MODELS REPORTED IN MANUSCRIPT glmmadmb(formula = citysexyselfies ~ zcitygi.health + zcitygi.reproductive + zcitygi.college + zcitygi.managerial + zcitygi.income + (zcitygi.health + zcitygi.reproductive + zcitygi.college + zcitygi.managerial + zcitygi.income citystate) + offset(log(citypopulation)), data = na.omit(subset(city, scale(city1.rirs$residuals, center = TRUE) <= 2.96 & scale(city1.rirs$residuals, center = TRUE) >= -2.96, select = c(citysexyselfies, zcitygi.health, zcitygi.reproductive, zcitygi.college, zcitygi.managerial, zcitygi.income, citypopulation, citystate))), family = "nbinom") AIC: (Intercept) <2e-16 *** zcitygi.health zcitygi.reproductive zcitygi.college zcitygi.managerial zcitygi.income ** Number of observations: total=5400, citystate=51 Group=cityState (Intercept) 2.450e zcitygi.health 1.858e zcitygi.reproductive 4.344e zcitygi.college 3.691e zcitygi.managerial 5.792e zcitygi.income 8.636e Negative binomial dispersion parameter: (std. err.: ) Log-likelihood:

29 FEMALE SEXUALIZATION ON SOCIAL MEDIA 29 City Model 2 glmmadmb(formula = citysexyselfies ~ zcitygini + (zcitygini citystate) + offset(log(citypopulation)), data = na.omit(subset(city, scale(city2.rirs$residuals, center = TRUE) <= 2.96 & scale(city2.rirs$residuals, center = TRUE) >= -2.96, select = c(citysexyselfies, zcitygini, citypopulation, citystate))), family = "nbinom") AIC: (Intercept) < 2e-16 *** zcitygini e-09 *** Number of observations: total=5513, citystate=50 Group=cityState (Intercept) zcitygini Negative binomial dispersion parameter: (std. err.: ) Log-likelihood:

30 FEMALE SEXUALIZATION ON SOCIAL MEDIA 30 City Model 3 glmmadmb(formula = citysexyselfies ~ zcitygi.health + zcitygi.reproductive + zcitygi.college + zcitygi.managerial + zcitygi.income + zcitygini + (zcitygi.health + zcitygi.reproductive + zcitygi.college + zcitygi.managerial + zcitygi.income + zcitygini citystate) + offset(log(citypopulation)), data = na.omit(subset(city, scale(city3.rirs$residuals, center = TRUE) <= 2.96 & scale(city3.rirs$residuals, center = TRUE) >= -2.96, select = c(citysexyselfies, zcitygi.health, zcitygi.reproductive, zcitygi.college, zcitygi.managerial, zcitygi.income, zcitygini, citypopulation, citystate))), family = "nbinom") AIC: (Intercept) < 2e-16 *** zcitygi.health zcitygi.reproductive zcitygi.college zcitygi.managerial zcitygi.income * zcitygini e-09 *** Number of observations: total=5398, citystate=51 Group=cityState (Intercept) zcitygi.health zcitygi.reproductive zcitygi.college zcitygi.managerial zcitygi.income zcitygini Negative binomial dispersion parameter: (std. err.: ) Log-likelihood:

31 FEMALE SEXUALIZATION ON SOCIAL MEDIA 31 City Model 4 glmmadmb(formula = citysexyselfies ~ zcitygi.health + zcitygi.reproductive + zcitygi.college + zcitygi.managerial + zcitygi.income + zcitygini + cityeduemplearnpca + zcitymedianagefemale + zcitysexratio + zcityurbanization + (zcitygi.health + zcitygi.reproductive + zcitygi.college + zcitygi.managerial + zcitygi.income + zcitygini + cityeduemplearnpca + zcitymedianagefemale + zcitysexratio + zcityurbanization citystate) + offset(log(citypopulation)), data = na.omit(subset(city, scale(city4.rirs3$residuals, center = TRUE) <= 2.96 & scale(city4.rirs3$residuals, center = TRUE) >= -2.96, select = c(citysexyselfies, citypopulation, zcitygi.health, zcitygi.reproductive, zcitygi.college, zcitygi.managerial, zcitygi.income, zcitygini, cityeduemplearnpca, zcitymedianagefemale, zcitysexratio, zcityurbanization, citystate))), family = "nbinom") AIC: (Intercept) < 2e-16 *** zcitygi.health * zcitygi.reproductive zcitygi.college zcitygi.managerial zcitygi.income * zcitygini e-07 *** cityeduemplearnpca * zcitymedianagefemale ** zcitysexratio zcityurbanization Number of observations: total=5375, citystate=51 Group=cityState (Intercept) zcitygi.health zcitygi.reproductive zcitygi.college zcitygi.managerial zcitygi.income zcitygini cityeduemplearnpca zcitymedianagefemale zcitysexratio zcityurbanization Negative binomial dispersion parameter: (std. err.: ) Log-likelihood:

32 FEMALE SEXUALIZATION ON SOCIAL MEDIA 32 County Models County Model 1 glmmadmb(formula = countysexyselfies ~ zcountygi.health + zcountygi.reproductive + zcountygi.college + zcountygi.managerial + zcountygi.income + (zcountygi.health + zcountygi.reproductive + zcountygi.college countystate) + offset(log(countypopulation)), data = na.omit(subset(county, scale(county1.rirs2$residuals, center = TRUE) <= 2.96 & scale(county1.rirs2$residuals, center = TRUE) >= -2.96, select = c(c ountysexyselfies, zcountygi.health, zcountygi.reproductive, zcountygi.college, zcountygi.managerial, zcountygi.income, countypopulation, countystate))), family = "nbinom") AIC: (Intercept) <2e-16 *** zcountygi.health zcountygi.reproductive zcountygi.college * zcountygi.managerial zcountygi.income * Number of observations: total=1588, countystate=51 Group=countyState (Intercept) zcountygi.health zcountygi.reproductive zcountygi.college Negative binomial dispersion parameter: (std. err.: ) Log-likelihood:

33 FEMALE SEXUALIZATION ON SOCIAL MEDIA 33 County Model 2 glmmadmb(formula = countysexyselfies ~ zcountygini + (1 countystate) + offset(log(countypopulation)), data = na.omit(subset(county, scale(county2.ri$residuals, center = TRUE) <= 2.96 & scale(county2.ri$residuals, center = TRUE) >= -2.96, select = c(countysexyselfies, zcountygini, countypopulation, countystate))), family = "nbinom") AIC: (Intercept) < 2e-16 *** zcountygini e-14 *** Number of observations: total=1601, countystate=51 Group=countyState (Intercept) Negative binomial dispersion parameter: (std. err.: ) Log-likelihood:

34 FEMALE SEXUALIZATION ON SOCIAL MEDIA 34 County Model 3 glmmadmb(formula = countysexyselfies ~ zcountygi.health + zcountygi.reproductive + zcountygi.college + zcountygi.managerial + zcountygi.income + zcountygini + (zcountygi.health + zcountygi.reproductive + zcountygi.college countystate) + offset(log(countypopulation)), data = na.omit(subset(county, scale(county3.rirs$residuals, center = TRUE) <= 2.96 & scale(county3.rirs$residuals, center = TRUE) >= -2.96, select = c(countysexyselfies, zcountygi.health, zcountygi.reproductive, zcountygi.college, zcountygi.managerial, zcountygi.income, zcountygini, countypopulation, countystate))), family = "nbinom") AIC: 4012 (Intercept) < 2e-16 *** zcountygi.health zcountygi.reproductive zcountygi.college * zcountygi.managerial zcountygi.income zcountygini e-13 *** Number of observations: total=1587, countystate=51 Group=countyState (Intercept) zcountygi.health zcountygi.reproductive zcountygi.college Negative binomial dispersion parameter: (std. err.: ) Log-likelihood:

35 FEMALE SEXUALIZATION ON SOCIAL MEDIA 35 County Model 4 glmmadmb(formula = countysexyselfies ~ zcountygi.health + zcountygi.reproductive + zcountygi.college + zcountygi.managerial + zcountygi. income + zcountygini + countyeduemplearnpca + zcountymedianagefemale + zcountysexratio + zcountyurbanization + (zcountygi.health + zcountygi.reproductive + zcountygi.college + countyeduemplearnpca + zcountymedianagefemale + zcountysexratio + zcountyurbanization countystate) + offset(log(countypopulation)), data = na.omit(subset(county, scale(county4.rirs2$residuals, center = TRUE) <= 2.96 & scale(county4.rirs2$residuals, center = TRUE) >= -2.96, select = c(countysexyselfies,zcountygi.health, zcountygi.reproductive, zcountygi.college, zcountygi.managerial, zcountygi.income, zcountygini, countypopulation, countystate, countyeduemplearnpca, zcountymedianagefemale, zcountysexratio, zcountyurbanization))),family = "nbinom") AIC: (Intercept) -1.25e e < 2e-16 *** zcountygi.health -2.15e e zcountygi.reproductive -2.06e e zcountygi.college -1.46e e zcountygi.managerial 9.78e e zcountygi.income -3.44e e zcountygini 2.68e e ** countyeduemplearnpca -1.69e e zcountymedianagefemale -2.02e e * zcountysexratio -4.77e e zcountyurbanization 9.51e e e-12 *** Number of observations: total=1535, countystate=50 Group=countyState (Intercept) 1.466e zcountygi.health 2.587e zcountygi.reproductive 8.703e zcountygi.college 1.018e countyeduemplearnpca 3.699e zcountymedianagefemale 8.275e zcountysexratio 3.343e zcountyurbanization 4.436e Negative binomial dispersion parameter: (std. err.: ) Log-likelihood: